POSSIBLE CONSTRAINTS ON ALLOPHONIC …ling.yale.edu/sites/default/files/files/alumni senior...1.1 The "Voicing" Contrastin Australia In discussing voicing contrasts, it is essential
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
POSSIBLE CONSTRAINTS ON ALLOPHONIC VOICING IN
AUSTRALIAN ABORIGINAL LANGUAGES:
EVIDENCE FROM BARDI, KAYARDILD, WARLPIRI, AND YAN-NHANGU
I could have never produced a momentous work like this without the support of faculty and
students of Yale University who have made this work possible. First, I would like to dedicate this thesis
to Professor Claire Bowern and postdoctoral associate Erich Round for supporting my interest and
research in Australian Aboriginal languages for the past two years. As specialists in Australian linguistics,
Claire Bowern and Erich Round were able to provide me with the data, the readings, and the theoretical
background necessary to pursue the mostly unexplored topic that I investigated in this thesis. As my
advisors Claire Bowern and Erich Round also had great insights into the issues I often grappled with, and, -in more ways than one, this thesis could not have proceeded without their oversight.
Many thanks also go to my LING 490/491 Professor Raffaella Zanuttini for her constant moral
support as Director of Undergraduate Studies in Linguistics during the course ofthis year. Along the
same vein, many thanks are in order for my fellow class of2010 Linguistics majors, who provided me not
only useful feedback on my topic throughout the year, but also through their constant encouragement
provided the inspiration to see my thesis to fruition.
I also owe infinite gratitude to two groups of people. First, I want to thank my family for
supporting my academic endeavors. Second, I want to thank the people and communities of Aboriginal
Australia (past, present, and future), without whose support the field of linguistics- and Australian
linguistics in particular- would be greatly impoverished.
Work on this essay was partially funded by NSF grant BCS-844550 "Pama-Nyungan and
Australian Prehistory", awarded to Claire Bowern and Yale University.
3
ABSTRACT
tentative until more data can be gathered and analyzed. ng IfJ/
Dorsovelar
k Ikl
m /n.1
It It Ity lei
ny /n /
Itl
n Inl
Consonants
d tal Apicoalveolar Lammopalatal RetroflexLamiaoden ,. "'- , u',=
th I! I
nh II) I
Bilabial
III Iml
p /pl
Nasal
Stop
languages (Evans 1996:724).
- hi h i itten in both practical Australianise and IPA orthography:follows, w lC IS wn
Aboriginal groups using 250 languages.
f h ld (Gaby 2008:212). These languages are divided between the Pama-the other languages 0 t e wor
~ ing roughly seven-eighths of the continent, and the Non-Pama-NyunganNyunganlanguages,occupy
A typical Australian phoneme inventory is exemplified in Kayardild as
compared to the other languages of the world,
, d h re up to seven distinct1 ' tori - typically lal Iii lui no fricative phonemes, an anyw esmall vowe mven ones ,
places of articulation, although many have five or six places of articulation (Butcher 2006: 190-191, Dixon
Prior to European contact, the Australian continent boasted about 7001980: 129-135, Bowem p.c.).
These Aboriginal languages developed in relative isolation from
1.0 INTRODUCTION
honolozi 1and phonotactic similaritiesThe Australian Aboriginal languages show remarkable p ono ogica
A number of these similarities are readily demonstrable and uniqueacross the Australian continent,
For example, many Australian Aboriginal languages have
Many Australian Aboriginal languages have a single phonological stop series, although others
have been reported to distinguish at least two stop series. The nature of the stop contrasts in Australian
languages has been variously described as that of voicing (Austin 1988), fortisllenis (Butcher 2004),
length (Evans & Merlan 2004), or even a combination of factors (Evans 1996). Few studies, however,
address the nature of allophonic stop variation that is also present throughout the Australian languages,
and even fewer studies attempt to compare phonetically the stops from languages across the Australian
continent. This work addresses these important gaps in the literature and attempts to shed insight onto
phonetics ofAustralian languages in general. To this end, several sound files from four representative
Australian languages were examined: Bardi, Kayardild, Warlpiri, and Yan-Nhangu. After segmenting
these files into individual words, I observed the phonetic context, broadly defined, in which the stops of
the language occur. Then I took several acoustic measurements: VOT, VTT, stop to release duration, and
total stop duration. The data appear to support the idea that stop allophony is contextually conditioned.
Furthermore the degree ofpersistent voicing in otherwise "voiceless" stops in the data may also confirm
the model ofpassive devoicing (Anderson & Maddieson 1994), However, these conclusions remain
Lateral /1/
Trill rr /rl
Approximant w I,vI y /]/ r /J I
front back
high i, i: u,u:
low a,a:
. dild N Pama-Nyungan languageTable 1. Phonemic Inventory of Kayar ) ,a on-
k! in the following sections of this work. TheI The ractical orthography will be used for the stops /p, th, t, ty, rt, _P di . d series will be written as follows: b, dh, d, dy, rd, g.correspon mg voice
45
l
As one can demonstrate, Kayardild possesses the typical three vowel contrast with a vowel length
distinction, six places of articulation, and does not have any fricatives in its inventory, although fricatives
do appear as variant forms of stops (Round 2009:52).
Kayardild also illustrates the general observation across Australian languages that there is only a
single stop series that typically does not contrast for voicing or duration. Hence if one is speaking
Kayardild one can in principle pronounce the word for 'woman' maku as either [maku] or [magu] with no
change in meaning, as /kJ and /g/ are not separate phonemes of the language. On the other hand, there
exist Australian languages with two stop series-perhaps up to 30%- where this generalization does not
hold (Butcher 2004:550). In such cases the voiced and voiceless stop series may in fact contrast
phonemically. Austin (1988) provides the example of the Gugu Uwanh minimal pairs illustrating such a
phonemic contrast: aku means 'skin', while agu means 'ground.'
The phonotactic patterns ofAustralian languages are also remarkable in that a typical word in an
Australian language is at least two syllables long. Thus a prototypical phonological word would be of the
form C1NlTIALVC1(C2)V(CF1NAL). The phonological voicing contrast typically restricted to the word-medial
C1(C2) position, although it can appear in other positions (Dixon 1980: 127). In addition, the word-medial
position is often the only position that allows consonant clusters (Butcher 2006:205). No Australian
language has a voicing contrast in all consonantal positions (Austin 1988, Butcher 2006:193).
1.1 The "Voicing" Contrast in Australia
In discussing voicing contrasts, it is essential to distinguish the phonetic aspect of "voicing,"
which is directly measurable to the acoustic signal, from its associated phonological dimension, which is
abstracted away from the signal, for example in setting up phonological contrasts. In the phonetic
literature, stop voicing is defined by the concomitant vibration of the vocal cords creating the relevant
glottal pulsing that is perceivable in the acoustic signal of the stop (Butcher & Reid 1989:6). From an
articulatory standpoint, this vibration arises from a pressure differential created by the lungs which forces
outgoing air through the narrow opening of the vocal cords. An absence of such concomitant vocal cord
6
vibration, for example in the case of the vocal folds being too far apart, thus characterizes the voiceless
stop (Ladefoged & Maddieson 1996:52). Voicing effects are measured in terms ofthe voice onset time
(VOT), defined as the time from the release of the stop to the resumption of regular voicing. In voiceless
stops, VOT corresponds to the length of the aspirated segment following the stop's release, whereas in
voiced stops, VOT corresponds to the duration of regular voicing prior to the release and is thus negative
(Ladefoged & Cho 200 I: 1). Voicing contrasts, in about half of the world's languages where the stop
contrast is phonological, employ different target VOT's for the stop distinction (Butcher & Reid 1989:6).
Although it is possible to differentiate voiced/voiceless stops by the maintenance of different
VOT's, it is not the case that these target VOT's are consistent either across or within languages. In an
investigation ofVOT across 18 languages, Ladefoged & Cho (2001) observed that human languages do
not appear to.converge on a common target VOT for each of their voiced and voiceless stops. Instead
much variation in VOT was observed, not only across languages as each language community settles on
its own target VOT values, but also within languages and across speakers as a result of other factors, like
Language Bilabial ~ Alveolar Retroflex Velar Uvular
Aleut (Eastern)59 75 78
Aleut (Western)76 95 92
Apache13 15 31
Apache (aspirated)58 80
BanawaH22
44
Bowiri17 18 39
Chickasaw 13 22 36
Dahalo20 15 42 27
Defaka18 20 30
Gaelic13 22
28
Gaelic (aspirated) 64 6573
1644 27
Hupa11
Hupa (aspirated)82
84
Jalapa Mazatec11 23
Jalapa Mazatec63 80
(aspirated) 20Khonoma Angami 10 9
Khonoma Angami 83 5591
(aspirated) 48 55
Montana Salish 22 24
Navajo12 6 45
Navajo (aspirated)130 154
Tlingit18 28 30
Tlingit (aspirated)120
128 128
Tsou11 17 28
Wari'19 26
50.58
Yapese20 22
56
Table 2. Mean VOT stop values showing considerable variation in VOT (Source: Ladefoged & Cho 2001)
7
II
I,
phonetic environment and natural variations in vocal tract physiology (Ladefoged & Cho 2001 :8). The
mean VOT values gathered by for the languages are reproduced in Table 2.
While Australian Aboriginal languages with two stop series can be analyzed as having a contrast
in voicing, recent phonetic research on Australian languages have called this assumption into question. A
handful of Australian languages do not consistently make a distinction in their target VOT's for their two
caveat. The phonetic work on Australian languages has already revealed that a lot of related phenomena
can be hidden in a simple phonological voicing distinction.
This present work will attempt to address allophonic voicing, a phenomenon that cannot be
understood from the phonological voiced/voiceless contrast model alone. It is traditionally assumed that
in most languages with a single stop series the stops are phonetically voiceless. Certainly this is the case
stop series (Butcher & Reid 1989:7). In fact, the phonological stop contrast may be linked to a
combination ofphonetic correlates, including voicing, stress, and vowel length (Evans 1996:730). For
example, the primary phonetic cue for the stop contrast for Jawoyn is stop duration. From acoustic data
in Polynesian languages like Hawaiian, Tongan, and Maori, where the stops all appear to be articulated
with open vocal cords (Ladefoged & Maddieson 1996:53). While Aboriginal Australian languages with
only one stop series typically have their stops described as voiceless and unaspirated, there are languages
languages, that would ensure the voicelessness of the stop, allowing residual voicing from the previous
These "voiced-stop" languages are remarkable until one considers another common pattern of
a:th
articulatory level, this may be due to the lack of an active laryngeal opening gesture, like in Polynesian
closure that fades into voicelessness. Such initial voicing may persist approximately 50 ms into the stop,
Wambaya, and Yuwaalaraay, are described as having regularly voiced stop allophones for its stop series,
with voiceless allophones sporadically manifesting in some contexts (Hamilton 1996:54).
segment to carry into the stop (Ladefoged & Maddieson 1996:54). An example of such a partially voiced
stop is illustrated in Figure 1, where the voicing from the vowel persists well into the dental stop.
completely voiced or voiceless, but rather "partially voiced." Such partially voiced stops usually are
recognizable word medially and have a common pattern of having a voiced portion at the onset of the
as reported in the Northern Territory language Tiwi (Anderson & Maddieson 1994:143). On an
where this generalization does not hold (Hamilton 1996:54). Some Australian languages, like Yidiny,
many Australian languages with a single stop series: many stops in these languages are not phonetically
in Jawoyn Evans & Merlan (2004:200) cite the word for 'frog' jatti as phonetically [jab], with the stop
possibly analyzable as a geminate. This is in contrast to the word for 'tributary' jateng, which is [jaden].
Yet this length distinction is not necessarily a perfect phonetic correlate of the phonological contrast, as a
certain degree of duration variability exists. For instance, the word 'spouse/in-law' kakkali is reported to
be pronounced either as [gagali] or [gak.ali] among Jawoyn speakers, although Evans & Merlan do not
cite the actual stop durations for these particular phonetic impressions (Evans & Merlan 2004:200).
Butcher (2004) argues for yet another model of the contrast: lenis/fortis. The lenis/fortis stop
distinction most famously appears in the study of various Korean dialects (see Cho et a12002). The
conceptual distinction between lenis/fortis is that fortis stops are produced with more respiratory effort
and tension than their lenis counterparts (Cho et al 2002: 194, Stoakes et al 2007:869). In contrast with
voicing and its VOT, however, the lenis/fortis contrast is not as well defined in terms of the phonetic
correlates, and this fact necessitates finding indirect phonetic correlates that hint at the lenis/fortis quality
of stops. From the acoustic and articulatory data in the Aboriginal languages Burarra, Murrinh-Patha, and
Bininj Gun-Wok, it can be argued that differences in the duration of the stop closure and peak intra-oral
pressure in the two stop series correlate roughly with the underlying "effort" and "tension" of the
lenis/fortis contrast in these Australian languages (Butcher 2004:556, Stoakes et a12007:869). Despite
the recent theoretical models of Australian stop contrast presented here, this work will, for convenience's
sake, continue the tradition of analyzing phonemically "voiced" and "voiceless" stops with one important
I
I,
i
8Figure 1. Partial Voicing is evident in the stop in Kayardild word muthaa (many)
II
I
III I
I,
I
With so much variation in the reported voicing of stops across Australian Aboriginal languages,
even in languages without the phonological voicing contrast, it is striking that little in the way of
comparative, instrumental analysis has been carried out to date on the allophonic voicing effects in
Australian languages. This present work hopes to address this gap in the literature and open up this field
of research to instrumental phonetic investigation and new insights on Australian Aboriginal languages.
1.2 Allophonic Voicing in Australian Languages: Hypotheses
Although allophonic voicing effects in Australian stops have often been reported in languages
with one stop series, the underlying cause of this allophony is currently not well understood. The
simplest explanation for this allophony is to assume that all allophonic forms of these Australian stops
occur in free variation. For instance, speakers of these languages can use anything available from [p] to
[b] for their labial stop and anything from [k] to [g] for their velar stop. Since the typical Australian
language also has no phonemic fricatives, one also has available the (typically voiced) spirantized
approximations [P] and [y] as other allophones ofthe labial and velar stop, respectively (Hamilton
1996:55). Hence, speakers can choose on arbitrary or extralinguistic grounds which stop allophones they
want to employ in their speech at any given moment. While of course there may be impressionistic
tendencies observed for certain allophones produced in certain contexts, these tendencies need not bother
one working under the free variation assumption. With some reservations, Dixon (1980) typifies this free
variation model in his remarks on Dyirbal pronunciation:
[diban], [tiban] are the two commonest pronunciations of 'stone.' [dipan], [tipan]
are heard much less often; but they are, unhestitatingly, taken as instances of the
same word since any allophone CAN be substituted for another allophone of the
same phoneme, without changing the meaning of the word. (127)
Free variation is certainly plausible in explaining the allophonic voicing in the Australian languages
without a voicing contrast. After all, Dyirbal speakers could in principle interpret with little difficulty the
words of another speaker who regularly alternated between [diban], [tiban], [dipan], and [tipan] forms,
10
and so in that sense the voicing of the stops in that word could be freely decided according to a speaker's
preference. In addition, it is likely that the context of the utterance can affect the stop's realization. For
example, a stop appearing in a word in citation form may very well vary from the same stop appearing in
continuous speech. However, if one assumes only free variation at work, then one cannot predict
anything about Dixon's "commonest pronunciations" of words with their target VOT's. After all, under a
free variation framework, one would predict that all allophones are permitted in all environments.
Another explanation for allophony is to assume that there exists an underlying normative
articulation (like voiceless unaspirated plosive) for voicing and manner for all the stops of Australian
languages. For instance, one can decide that [p, b, P] are all representations of the stop that is typically [P]
and [k, g, y] are all representations of the stop that is typically [k], with the allophonic variants arising
from some phonologically derived environments: for example, we can suppose that voiced stops surface
only in an intervocalic environment, like the [b] in Dixon's [tiban] example. While this hypothesis
creates testable predictions on allophonic distribution, like which allophone is most likely to surface
phonetically in what environment, several remarks remain to be made regarding this normative
articulation hypothesis. It is of course premature to assume that for Australian languages the normative
stop articulation is always the voiceless unaspirated plosive allophone. In languages with most stops
phonetically voiced, as in Yidiny, we would likely draw the conclusion that in such a language the
allophones [p, b, P] would all represent [b] instead of [p]: Each language in Australia would be able to
choose their normative articulation in this way.
A related approach to explain allophonic voicing in Australian languages with one stop series is
to assume that these stops have no "underlying" voice value and thus are phonologically underspecified
for voicing. According to a phonological analysis based on underspecification, the realizations of the
stops would not be arbitrary, with the voicing parameter largely dependent on either universal or language
specific principles. For example, if one observes that Australian stops tend to voice in intervocalic
position, then one can formulate the rule: [+stoP] ~ [+voice] / V_V. While it is possible to make precise
11
(laminopalatal /ty/) and lack the laminodental stop /th/.
Of these four languages, only Yan-Nhangu is described to have a phonological voicing contrast
that while Kayardild and Yan-Nhangu have all six places of articulation /p, th, t, ty, rt, k/ for their
Knynrdild
Figure 2. Geographicdistributionoflanguagesinvestigated in this work
languages with only one stop series. After all, this language should be expected to exhibit less variation
(Baymarrwanga et aI2008:34). Yan-Nhangu thus can serve as a control case to compare against the
in the voicing parameter compared to the languages without a voicing contrast, since voicing is
phonemic stop inventory, Bardi and Warlpiri only have five, as they only have a single laminal stop
Nyungan languages, whereas Bardi and Kayardild are Non-Pama-Nyungan. It is also interesting to note
While a detailed Australia-wide survey of
four languages have individual differences of interest. For example, Yan-Nhangu and Warlpiri are Pama-
because ample phonetic data was available to investigate these languages in depth. Nevertheless, these
languages were far from being representative of the Australian continent, and they were chosen primarily
work, the languages under investigation consisted of Bardi, Kayardild, Warlpiri, and Yan-Nhangu. These
across the Australian continent as possible. In this
linguistic, phonological, and phonotactic variety
Australian languages that are as representative of the
of allophonic voicing effects to choose a sample of
this work, it was of great interest for the investigation
languages would be too ambitious for the scope of
2.0 RESEARCH METHODS
From the previous discussion and the literature it encompasses, it is clear that there is still much
A final approach for allophony is the passive devoicing model. In this model, the stops in
Australian languages are produced with no actual laryngeal opening gesture, so the voicing from the
previous voiced segment, like a vowel, can passively persist well into the start of the stop until it fully
devoices due to passive cavity expansion (Ladefoged & Maddieson 1996:54). Hence, the continuum of
voiceless to voiced stops is a direct product of voicing duration from the preceding segment carrying over
into the stop. This model, which relies primarily on the phonetic environment and intrinsic phonetic
properties of stops, creates many testable predictions about the distribution of allophonic forms. For
example, we can predict that the maintenance of stop voicing in such a model should be affected by both
stop duration and place of articulation. In particular, it is harder to maintain voicing passively through a
longer stop, like a geminate, than a shorter stop, because the pressure differential provided by air escaping
through the vocal cords can only be maintained for so long before the air supply maintaining voicing runs
out. In addition, it is harder to keep a stop articulated at the back of the mouth, like a 1kI, voiced than a
stop articulated at the front, like a /p/, because of the differing sizes of the cavity behind the oral
constriction (Hayes & Steriade 2004:12-13).
predictions like this one according to the rules of underspecification, an inadequacy with this framework,
and other phonological approaches to allophonic voicing, is that the phonetic reality ofAustralian stops is
often more complex than a binary [±voice] parameter that needs to be filled. In particular, these quasi
phonological approaches have little to say about the "partial voicing" effects on the phonetic level
reported in many Australian languages.
,
! I
I'
! I
to be learned and refined from the study of allophonic voicing effects in Australian languages. Certainly
none of these models of allophonic voicing need to be rejected a priori, and none of these models are
phonologically contrastive in Yan-Nhangu.
mutually exclusive. This present work will consider the hypotheses presented here using data from
several Australian languages and provide some insight as to some possible phonetic constraints to such
stop allophony.
2.1 Speakers and Sources
Utterances from speakers of these four languages were gathered from various sound files from
field recordings and leamer's guides. Both the field recordings and the leamer's guides contained a
collection of directly elicited forms as well as dialogues in the respective languages. The audio files were
Table 4. Examples of word tokens containing the velar stop sorted by preceding segment values
The stop's syllabic position was another independent variable investigated in this work, coded as
Cl, C2, C3, etc., where Cl represents any initial consonant onset joined to the first syllable, C2 represents
any consonant associated with the second syllable, and so on. Hence, the velar stop in the Bardi words
agal 'and' and dalga 'cowrie shell' were coded as C2, whereas bilanggamarr 'helicopter tree' was coded
as C3. In this work, syllabic position was an indirect indicator of the degree of stress of the syllable that
may affect the quality of the stop. In all languages in the sample, stress was always initial and associated
with the syllable associated with Cl position. Thus, C2 always described the syllable associated with
post-stress position, and C3 always described the syllable associated with the next position, and so on.
2.2.2 Dependent Variables: VOT, VTT, Stop Closure to Release, etc.
Each sound file containing a word token was individually labeled with 5 interval tiers in Praat.
The measures of interest taken included the entire duration of stop closure to the release of the articulators
17
(see I at Figure 6), voice onset time (VOT): the duration from the release of articulators to the resumption
of regular voicing (2 at Figure 6), and voice termination time (VTTi: the duration of the transition state
from the end of regular voicing up to the final closure of the articulators (3 at Figure 6). Durations were
measured visually using time-aligned waveform and spectrogram windows. Other measures that were
taken include the duration of the preceding segment and the total stop duration (calculated by summing up
I and 2 in Figure 6). As mentioned earlier, many variables could affect the durations measured in this
study: preceding segment, syllabic position, speaker variation, and so on.
in a voiced stop is never supposed to "terminate." However,because a voiced stop can be approximated
as a stop with the maximum VTT value, the VTT can thus be assumed out of convenience to be equal to
the closure duration.
The phonetically spirantized allophones of stop tokens, because of their tendency to be voiced,
were treated similarly to voiced stop tokens, although these fricative allophones have neither VOT nor
VTT values. Spirantized allophones seemed to occur in all environments in free variation with the
corresponding stops (in Yan-Nhangu the voiced stops in particular), although the voiced velar
fricative [y] was the most commonly observed across the languages studied. The notable
exception in this discussion is in Kayardild, where no instances of spirantization were recorded.
Table 5 summarizes the occurrences of fricative stop allophones in this study.
ng a 3;,,, 3.0 RESULTS
Figure 6. Example of stop-related measurements for Warlpiri ngaka 'later', modeled after Stoakes et aI, 2007.
While the measurements of parameters like VOT and VTT in voiceless and partially voiced stop
tokens were straightforward, two special types of tokens deserve special consideration. For the
phonetically voiced stop tokens, the VOT is always negative, and the VTT is ill-defined, since the voicing
3 VTT specifically measures what this workhasbeenreferring up to thispointas the extent of "partialvoicing"noted in Australian stops.
18
3.1 Preliminary Investigation of Syllabic Position
Many of the stop tokens observed were in the post-stress C2 position, although a few were in C3
or other positions. This bias towards C2 position tokens was out of convenience, as it was easiest to find
examples of stops and phonetic environments in C2 position. As a result, while it was not possible to
explore in depth how much of a confounding role syllabic position played in all the languages studied in
this work, it was at least feasible to lookat-the.effect syllabic position may play in Warlpiri. After all,
19
there were enough C2 and C3 tokens (primarily for Ip/, Ity/, and 1kI) to make preliminary observations on Warlpiri Stop Duration by syllabic position300 ,-'--------------------~
the effect of syllabic position on this study's dependent variables. VTT, VOT, and stop duration were all *250
compared across C2/C3 positions in Warlpiri, as illustrated in Figures 7,8, and 9. The overall clustered
distribution of these values with respect to C2 and C3 show that the effect of C2 vs. C3 syllable position
on the dependent variables was minimal at best, at least in Warlpiri. However, this result does not rule
out the possibility of measurable effects in other syllabic positions, like the stressed Cl position.
Table 6. Tokens sorted by language and voicing parameter. For the sake of discussion, partially voiced stops aregrouped with the other voiceless tokens.
20 21
3.2.1 Mean VTT duration across place ofarticulation /k/, are harder to keep voiced than stop segments formed in the front of the mouth, like bilabial /p/ (Hayes
Mean VTT durations for voiceless stops seemed to vary according to language. For example, in & Steriade 2004: 13). Hence, one would predict that there would be a negative correlation between the
Kayardild the average VTT value across all voiceless or partially voiced stops was around 60 ms, whereas backness of the stop constriction (from /p/ to /k/) and the average VTT duration.
Warlpiri and Yan-Nhangu's voiceless stops had average VTT durations of around 30 ms. The stops In order to test this hypothesis, the Pearson's correlation was calculated for Kayardild, Warlpiri,
appeared to vary less as a function of place of articulation, however. This could suggest a common and Yan-Nhangu's voiceless stops. While in Warlpiri and Yan-Nhangu a negative correlation was found
underlying process, like passive devoicing, that is active across all places of articulation which results in (Walpiri r = -0.06, YN r = -0.24, compared with Kayardild r = 0.022), only in Yan-Nhangu did this
approximately consistent VTT averages, in a manner similar to the VTT averages occurring in Tiwi correlation find statistical significance with p < 0.05 (YN p = 0.008). This lack of a negative correlation
across place of articulation (Anderson & Maddieson 1994:143). In the case ofYan-Nhangu, which does
contrast voiced and voiceless stops, the shorter average VTT duration (resulting in less partially voiced
stops) might reflect the tendency for stops in this language to devoice sooner in order to make the
in Warlpiri was not surprising, given the outliers in VTT values across each place of articulation for
Warlpiri as shown in Figure 10. If one compares the VTT distribution ofKayardild and Warlpiri with
that of Yan-Nhangu (Figure 11), it becomes evident that by inspection that Yan-Nhangu's stops had a
voiced/voiceless contrast more salient. Table 7 summarizes the average VTT durations across the various more defined downward sloping trend from /p/ to /k/ than either of the other two languages.
22
If the passive devoicing model indeed applies to these stops, then it should be possible to predict
- Mean 31.92
**
*
*
rt tilJVarlpiri
Warlpiri VTT
180
160
140
120
i 100
:: 80>
60
40
Mean 27.5320
0StDp t rt ty
Language Yan-hJhangu
Figure 11. Boxplot ofVTT (ms) according to place of articulation in Yan-Nhangu.
---~3\
)
Yan-Nhangu (Voiceless) VTT
Kayardild VTT
180 180
160 160
140 140
120 120
" 100 i 100.§.>- *>- 80 :: 80> >
60 Mean 63,9560
40 40
20 20
0 0Stop th rt ty StDp
Language Kayardild Language
Figure 10. Boxplot ofVTT (ms) according to place of articulation in Kayardild and Warlpiri.
20.5527.53124
YN (voiceless series)
24.1531.92
Warlpiri
18630.0363.95
Kayardild
142Sum
from the phonetic literature it is known that stops with constrictions in the back of the mouth, like a velar
the subtle effect ofplace of articulation on the duration of persistent voicing measured by VTT. After all,
Table 7. Average VTT durations (ms) across place of articulation from three Australian languages. Bardi and YanNhangu's voiced stops are excluded because their mostly voiced stop tokens lack comparably defined VTT values.
places of articulation for Kayardild, Warlpiri, and Yan-Nhangu.
P 30 65.04 32.74 P 40 38.61 19.52 P 28 31.82 24.16
th 25 65.06 23.03 th - - - th 22 25.41 19.04
t 19 75.28 38.27 t 34 27.82 15.85 t 12 43.00 25.47
ty 22 65.06 31.19 ty 40 33.39 31.08 ty 16 29.94 18.47
k 34 63.06 32.74 k 52 31.20 28.74 k 33 16.02 13.25
i i
3.2.2 Mean VOT duration across place ofarticulation3.3 Averaged Durations: Models
Stop components (2) (3) (4)
Preceding Segment (1)
(4)
(3)
(2)
(1)
(1) Average Duration of Preceding Segment
(2) Average VTT
(3) Average Voiceless Portion of Closure
(4) Average (positive) VaT
(4)(3)
,. ,, ,, ,, ,, ,
~2), ,, ,, ,
(1)
Actual Waveform & Spectrogram Average Stop (Abstracted from 4 measurements)
25
Figure 13. Mean ~eas~rements .abstracted from the stop and its immediate environment (preceding segment).The average VOT III this model IS not represented here if it is negative.
While mean VOT and VTT measurements are certainly insightful, they do not summarize in
isolation enough of the relevant phonetic context for the purposes of this work. Thus it was necessary to
abstract away an "averaged stop" from the data combining the relevant mean measurements obtained in
this study (see Figure 13). In particular, this work was interested in comparing the stops across languages
for stop category and the nature of the preceding segment- vowel (V), liquid (L), rhotic (R), nasal (N).
The average stops were generated for each of the individua11anguages studied and will be discussed in
further detail. The parts of the average stop are be abbreviated as (1), (2), (3), and (4). For the purposes
of discussion, average stop duration does not include the duration of the preceding segment.
15.5427.37123
YN (voiceless series)
17.1735.56
Warlpiri
1289.7525.39
Kayardild
142Sum
Warlpiri VOl Yan-Nhangu (Voiceless) VOl
90 90
80 80
*70 70
*60 *
60 *"' 50 "' 50 *-5. -5..... ....o 40 o 40> - Mean 35.56 >.
ty 15 30.60 11.81 ty 33 53.36 14.47 ty 16 33.81 12.82
k 28 29.38 7.66 k 32 35.73 12.26 k 33 29.85 22.23
Mean VOT values were more variable within the stop categories than across languages. On
24
establish any statistical significance in VOTs between all the stops.
considerably aspirated with the longest VOT value, and the retroflex stop (rt) less so with the shortest
around 30-40 ms (see Table 8). Within Warlpiri and Yan-Nhangu, the voiceless palatal stop (ty) was
average, the mean VOT values for all stops across the languages studied in this work were on average
be differentiated by VOT cues, although there were not enough tokens in the analyzable data set to
VOT value (see Figure 12). These observations suggest that the placement of individual stops may in part
Figure 12. Boxplot ofVOT (ms) for stops across place of articulationfor Warlpiri and Yan-Nhangu. Note thedifference in VOT distribution betweenthe retroflex (rt) and palatal (ty) stops.
Table 8. AverageVOT values for stops across place of articulation from three Australian languages,with the VOTvalues from Bardi and Yan-Nhangu's voiced stops again excluded, as well as any other voiced stops in the otherthree languages.
3.2.1 Bardi
In Bardi, a language with its stops typically phonetically voiced, it was expected that its (3) and
(4) values contributed little to the averaged stop, and that the stop would consist solely of persistent
voicing duration (VTT), which for voiced stops was defined as equal to the closure duration. This is what
was in fact generated in Figure 14. An interesting pattern also emerges in Bardi: the duration of the stop
seemed to be systematically correlated with the sonority of the preceding segment. For example,
regardless of place of articulation, stops preceded by the more sonorous vowel appeared to be longer in
duration than liquids (laterals and rhotics) which in turn are longer than stops preceded by nasals. This
trend did not apply only to Bardi but to other languages as well. The durations of the average stops
ranged anywhere from about 25 to 125 ms depending on place of articulation and preceding segment.
stop consisting of a voiceless segment (3) and positive VOT (4) duration. Like in Bardi, stops that were
preceded by the more sonorous vowel tended to be longer than those preceded by liquids which were in
tum longer than those preceded by nasals. Perhaps as a natural consequence of this correlation between
preceding segment sonority and duration did stops in nasal-stop clusters tend to be more voiced than in
other environments, as has been observed by Hamilton (1996: 54). After all, if one assumes a relatively
constant VTT value across all stops due to the mechanics ofpassive devoicing, and the stop is relatively
shorter in a nasal-stop cluster, it is likely ceteris paribus that residual voicing would persist through the
stop than in the case of a longer stop (Ladefoged & Maddieson 1996:53).
One interesting pattern in Kayardild was the tendency for many apical clusters" to have a trilled
release (Round p.c.). For example, for the Kayardild word token kultu 'flank,' the liquid-stop cluster was
pronounced somewhat like [kuld'u]. This trilled release posed methodological problems to the stop
Figure 14. Average stop durationfor Bardi (ms). The tokens with preceding segmentsas long vowelswere excluded
from the average stop analysis.
duration labeling process, since the trilled release would occupy the place of the normal aspirated VOT
• AvgVOT
• AvgPrecSegment
• AvgVoiceless Closure
I I I 1 -I I I I I I I I I I Ip th rt tv k
50
o
-50
150
100
-100
250
200
segment (4) of a typical stop and was thus coded for such in the average stops, although it is unclear
Butcher, Andrew. 2004. 'Fortis/lenis' revisited one more time: the aerodynamics of some oral stopcontrasts in three continents. Clinical Linguistics & Phonetics 18:6. 547-557.
Butcher, Andrew & Nicholas Reid. 1989. Redressing the balance on Australian stop contrasts:Comments on Austin's (1988) "Phonological voicing contrasts in Australian aboriginallanguages.' Canberra, ACT: Australian National University manuscript.
Butcher, Andrew. 2006. Australian Aboriginal Languages: Consonant-Salient Phonologies and the 'Placeof-Articulation Imperative'. In Speech Production: Models, Phonetic Processes, and Techniques,187-210. New York, USA: Psychology Pres.
Dixon, Robert M.W. 1980. The languages of Australia. Cambridge: Cambridge University Press, 125194.
Cho, Taehong & Sun-Ah Jun & Peter Ladefoged. 2002. Acoustic and aerodynamic correlates of Koreanstops and fricatives. Journal ofPhonetics 30. 193-228.
Evans, Bethwyn & Francesca Merlan. 2004. Stop Contrasts in Languages ofArnhem Land: From thePerspective of Jawoyn, Southern Arnhem Land. Australian Journal of Linguistics 24(2). 185-224.
Evans, Nick. 1996. Current Issues in the Phonology of Australian Languages. In John A. Goldsmith(ed.), The Handbook ofPhonological Theory, 723-759. Cambridge, MA & Oxford: Blackwell.
Gaby, Alice. 2008. Rebuilding Australia's Linguistic Profile: Recent Developments in Research onAustralian Aboriginal Languages. Language and Linguistics Compass 2(1). 211-213.
Hamilton, Philip James. 1996. Constraints and Markedness in the Phonotactics of Australian Aboriginallanguages. Toronto, CA: University of Toronto PhD thesis. '
Hayes, B. & Steriade, D. 2004. Introduction: the phonetic basis of phonological markedness. In BruceHayes, R. Kirchner & Donca Steriade (eds), Phonetically-Based Phonology, 1-33. Cambridge:Cambridge University Press.
Laughren, Mary & Robert Hoogenraad & Kenneth Hale & Robin Japanangka Granites. 1996. ALeamer's Guide toWarlpiri: Tape course for beginners. Alice Springs, NT: lAD Press.
36
Ladefoged, Peter & Taehong Cho. 2001. Linking linguistic contrasts to reality: The case ofVOT. In N.Gronnum & J. Rische1 (eds.), Travaux Du Cercle Linguistique De Copenhague, 212-225.Copenhagen: C.A. Reitzel.
Ladefoged, Peter & Ian Maddieson. 1996. The Sounds of the World's Languages. Oxford, UK:Blackwell Publishers, Ltd, 47-57.
Round, Erich. 2009. Kayardi1d Morphology, Phonology, and Morphosyntax. New Haven, CT: YaleUniversity PhD thesis.
Stoakes, Hywe1 & Janet Fletcher & Andy Butcher. 2007. An Acoustic Study of Bininj Gun-Wok MedialStop Consonants. International Congress ofPhonetic Sciences (ICPhS 2007 Germany) 16. 869872.